Marco Peereboom <slash <at> peereboom.us> writes:

> The plugging in of the disk is a non-event.  The disk is dead to the
> OS and by extension to softraid.

Let me follow up on this topic, please, and report some more experiments and
results and thoughts.

I recreated the mirror from scratch, and put /tmp, /var, /usr, /home and 
/backup directories on it. (No need to point out this is kind of stupid.)
Running for 2 days. 
Hot unplugged drive A. Then 'echo Nonsense > /backup/testo'
Good outcome, though not tested intensely yet: the system keeps running on 
B as if nothing had happened. 
Shutdown and plugged A back, restart.
Fails at file check, with 'help!' and dropping to a shell at /var. 
Problem is, that the .pid had been properly removed on B, but not on A; 
and I needed to delete those one by one at fsck. I also fsck-ed all other
partitions, and as to be expected, the 'testo' was on B, not on A, 
and therefore it needed to be deleted.
Reboot, alas, ending in a hangman. Reboot.
Another time /var drops to a shell, it has some trouble with 'lost+found',
another manual fsck is needed, reboot.
Finally, the mirror comes up properly. 

Next, I'd like to do a real test on a production machine. What scares me, 
is the lack of physical access, so the hangman and the drop to shell for 
fsck are not good. And, on a production box here, there might be thousands 
of files accumulating on the plugged drive that won't be available on the 
unplugged one,
and I will be asked to delete those. Also, this is not good. 
My question/suggestion: I for one would be happy if the state after reboot 
would by default be identical to the (degraded) state before the reboot: 
Because then I would hope to get the system started without the earlier 
defunct drive; that means, hopefully starting okay, and more relevant, 
not require me to do anything, not to delete any files. Simply start with 
the sane drive of the broken mirror as it was shut down. Then I could 
dump and restore the data to a freshly created RAID, without any further ado.
Then, at least, a broken drive, a flimsy controller would not interfere 
into the proper running and restarting of the box; and giving me the 
chance to retrieve all, including the most recent, data.

Does this make sense?

Uwe

Reply via email to